Skip to main content
Sonian is powered by modern large language models (LLMs). As powerful as LLMs are, they can sometimes produce inaccurate or out-of-scope answers. This is called “hallucination”. Sonian provides multiple mechanisms/features to decrease hallucination.

​
Avoid duplicated content

Adding multiple times the same documents/information, even translated should be avoided. Duplicated content can impact your agent’s performance. It is best to add a piece of information only once, whatever the language

​
Use a stronger model

Using a more capable model can improve answer quality and instruction following. The tradeoff is usually higher cost and slightly slower response times.

​
Fine-tune your Agent’s answers

Use our latest feature, Fine-tune Answers, to correct an Agent’s answer.